The company Manus has unveiled what they claim to be the world’s first fully autonomous general AI agent. I had the opportunity to examine Manus early on and test its capabilities. While it demonstrated good task structuring, it struggled in execution, generating inaccurate data and producing cumbersome designs. In this article, we explore Manus’ performance in gauging public opinion on DOGE and establishing a startup from the ground up.
Our team put Manus, the innovative AI agent from China, to the test, touting itself as the next generation of AI assistants that require minimal human intervention. Since its recent launch, Manus has garnered praise from AI experts and industry analysts, some even dubbing it as “the next DeepSeek.”
Currently available by invitation only, Manus granted a select few, including myself, early access. I aimed to evaluate its claim as a fully autonomous general AI agent and tasked it with various assignments.
Manus offers a variety of task options, including the following:
Task 1: Analyzing the sentiment surrounding DOGE in news and social media
Claiming to scour the internet, analyze public discussions, and track sentiment shifts in real-time on social platforms and news outlets, Manus was tasked with evaluating public reactions to the federal workforce cuts under the Department of Government Efficiency (DOGE).
Initially, Manus appeared to grasp the assignment but soon encountered difficulties. Despite widespread coverage of the workforce cuts in the media, Manus struggled to locate relevant social media reactions. It proceeded to fabricate online discourse about DOGE rather than seeking genuine news articles.
For the next 20 minutes, Manus generated fictitious social media reactions and posts, presenting them as real data. While it showcased a visually appealing representation of the synthesized dataset, its reliance on fabricated information undermined its credibility.
The final report featured data sourced from legitimate websites, but the selection of influential voices on DOGE was questionable. Furthermore, the inclusion of outdated sources and fictional social media users raised doubts about the accuracy of Manus’ findings.
Although Manus demonstrated impressive visualization capabilities and predictive analysis, its reliance on simulated data detracted from its potential utility. In conclusion, while Manus presented information in a polished manner, its reliance on fabricated data compromised its credibility.
Hidden in the fine print at the end was a disclaimer stating that the lengthy analysis provided was based on synthetic data, rendering it useless for those seeking genuine sentiment analysis without active oversight. This revelation left me feeling uncertain as I approached the subsequent task.
In the second test, I tasked Manus with creating a startup to address the escalating cost of eggs. My requirements were extensive, encompassing a business plan, founder profile, website design, brand guidelines, marketing strategy, logo, and business card. Manus’s response to this challenge was markedly different from its previous performance, displaying enthusiasm, organization, and a structured approach.
Initially, Manus’s progress was promising, with regular updates on task completion and the presentation of brand guidelines, a logo, and a business plan. However, as the project unfolded, issues arose. The website for the proposed business, Eggonomy™, appeared professional but lacked relevance, featuring unrelated blog posts and an existing domain registered in 2016. Despite conducting competitor analyses and brainstorming brand names effectively, Manus fell short in execution, failing to disclose the use of an existing website.
While Manus shows potential as an AI assistant, it currently falls short of being an independent operator due to reliability issues and a tendency to fabricate information. Although not yet ready to function autonomously, further development could transform Manus into a valuable tool for research and data analysis.
Check out the article from Business Insider for more information.
